Update app.py

app.py

@@ -12,15 +12,19 @@ import torch
 processor = Wav2Vec2Processor.from_pretrained("jonatasgrosman/wav2vec2-large-xlsr-53-italian")
 model = Wav2Vec2ForCTC.from_pretrained("jonatasgrosman/wav2vec2-large-xlsr-53-italian")
 
+def preprocess_audio(audio_data):
+    # Apply any custom preprocessing to the audio data here if needed
+    return processor(audio_data, return_tensors="pt").input_features
+
 # Function to perform ASR on audio data
-def transcribe_audio(audio_data):
-    print("Received audio data:", audio_data)  # Debug print
+def transcribe_audio(input_features):
+    print("Received audio data:", input_features)  # Debug print
 
     # Check if audio_data is None or not a tuple of length 2
-    if audio_data is None or not isinstance(audio_data, tuple) or len(audio_data) != 2:
+    if audio_data is None or not isinstance(input_features, tuple) or len(input_features) != 2:
         return "Invalid audio data format."
 
-    sample_rate, waveform = audio_data
+    sample_rate, waveform = input_features
 
     # Check if waveform is None or not a NumPy array
     if waveform is None or not isinstance(waveform, torch.Tensor):
@@ -29,10 +33,10 @@ def transcribe_audio(audio_data):
     try:
         # Convert audio data to mono and normalize
         audio_data = torchaudio.transforms.Resample(sample_rate, 100000)(waveform)
-        audio_data = torchaudio.functional.gain(audio_data, gain_db=5.0)
+        audio_data = torchaudio.functional.gain(input_features, gain_db=5.0)
 
         # Apply custom preprocessing to the audio data if needed
-        input_values = processor(audio_data[0], return_tensors="pt").input_values
+        input_values = processor(input_features[0], return_tensors="pt").input_values
 
         # Perform ASR
         with torch.no_grad():
@@ -48,5 +52,5 @@ def transcribe_audio(audio_data):
         return f"An error occurred: {str(e)}"
 
 # Create Gradio interface
-audio_input = gr.Audio()
+audio_input = gr.Audio(sources=["microphone"])
 gr.Interface(fn=transcribe_audio, inputs=audio_input, outputs="text").launch()